基于标记和模糊聚类的分水岭声纳图像分割  

Segmenting watershed sonar image by marker and fuzzy clustering

作　　者：李轲[1] 刘忠[1] 李翀伦[1] 张国栋[1] 

机构地区：[1]海军工程大学电子工程学院,湖北武汉430033

出　　处：《华中科技大学学报（自然科学版）》2013年第6期50-54,共5页Journal of Huazhong University of Science and Technology(Natural Science Edition)

基　　金：高等学校博士后专项科研基金资助项目(20090641460)

摘　　要：针对传统分水岭算法在处理声纳图像时存在严重的过分割现象,提出一种结合分割前处理和后处理两类方法优点的算法.首先利用H-min变换技术提取区域极小值和新的标记,对标记后的图像进行分水岭图像分割;然后结合改进适应度函数的粒子群全局寻优算法,从初分割的小区域中搜索出较为准确的初始聚类中心,利用这个聚类中心和改进目标函数的模糊C-均值聚类算法,再对分割后的小区域聚类,并控制迭代次数,以提高分割速度.实验结果表明:该方法能够有效消除过分割现象,提高声纳图像处理效果,有效分割率达89%,处理时间提高30%以上.To solve lng sonar image, a the problem of overs-egmentation in traditional watershed algorithm when process watershed segmentation algorithm based on marker and fuzzy C-means clustering （FCM） was proposed. The H-rain transform was firstly used to pick up the image region minimum and new marker. Then the marked image was segmented by watershed algorithm. After that, the im- proved particle swarm optimization （PSO） was used to find the accurate original elustering centers of FCM. With the help of these centers and the improved target function, the small regions of the initial segmented image were clustered by FCM. The iterating number was controlled to increase segment speed. The experimental results show that this method can solve the problem of over-segmentation and increase the sonar image segmentation efficiency with the effective segmentation rate of 89 ~ and the process time of more than 30 %increase.

关 键 词：图像分割 声纳图像 分水岭算法 标记 粒子群寻优 模糊C-均值聚类 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]